Apprehensions regarding the accuracy of atrial fibrillation (AF) detection by smartwatch algorithms persist despite their growing popularity. As false positive AF alerts can negatively impact a patient’s perceived physical wellbeing and confidence in managing chronic symptoms, understanding the potential sources of algorithmic error is crucial for informed patient care.
Smartwatch ECGs are acquired in uncontrolled environments, subjecting these signals to less-than ideal recording conditions. This, paired with the unique ECG morphology across a real-world population, increases the likelihood for false results, which when received by a lay person, may raise stress levels and health anxiety.
The primary aim of this study was to identify limitations in smartwatch acquired ECGs including their generalisability across a real-world population, highlighting the importance of independent validation through the HeartKey Rhythm Classification algorithm when assessing the reliability of consumer ECG data.
A 38-year-old male with intermittent palpitations, clamminess, and lightheadedness developed anxiety after receiving multiple ‘Atrial Fibrillation’ (AF) alerts from his smartwatch ECG (Samsung® Galaxy Series 4). Prompted by an ‘unusual sensation in the heart’, a familial history of MI, and a perceived worsening of symptoms, the patient presented at the Emergency Department, where a 12-lead ECG revealed a normal sinus rhythm of 90 bpm. The patient was reassured, discharged in a calm state, and referred for cardiological consultation. Holter monitoring on three subsequent occasions, echocardiographic examination, and coronary CT all proved unremarkable. Despite this, the patient continued to receive false AF triggers, resulting in ongoing anxiety and health issues. A troubleshooting session confirmed proper device use and revealed that all AF alerts were false, with the patient actually in sinus rhythm. Notably, the erroneous interpretations did not align with common sources of algorithmic error, such as artefact or ectopy, and manifested in signals of both high and low quality. A consistent feature was the patient’s atypical waveform which deviated from a typical smartwatch signal resembling Einthoven Lead I. To provide reassurance, the false positive data was extracted and re-analysed using HeartKey®, which correctly identified signals as either sinus rhythm or inconclusive.
This case study raises questions around the universality of AF detection on consumer devices such as smartwatches and underscores the need for improved clarity and accountability in the design and deployment of such technologies. The importance of independent validation of consumer ECG data is highlighted through results returned by HeartKey.